Explore Projects

In the current research ecosystem, datasets and trained machine learning models (e.g., LLMs, CNNs, etc.) are often scattered across various repositories and platforms, making discovery and reuse inefficient. This project proposes the design and development of a unified user interface system within Apache Airavata to consolidate and manage both datasets and models in one accessible platform. The contribution is aligned with the ticket: Come up with UX designs for Airavata data and model catalog. Through modern UI/UX design and seamless integration, this system will allow researchers and developers to find, preview, download, and reuse datasets and AI models effortlessly.

<p>Using Deep Learning techniques in order to model the relationship between the planetary albedo and chemical composition of Mercury.</p>

Xeus-cpp, a Jupyter kernel for C++, is built upon the native implementation of the Jupyter protocol, xeus. This setup empowers users to interactively write and execute C++ code, providing immediate visibility into the results.With its REPL (read-eval-print-loop) functionality, users can rapidly prototype and iterate without the need to compile and run separate C++ programs. Additionally, this integration facilitates seamless interaction between C++ and Python within the same Jupyter environment. With the ever growing popularity of large language models, this project aims to integrate a large language model with the xeus-cpp Jupyter kernel. This integration will enable users to interactively generate and execute code in C++ leveraging the assistance of the language model.

<p>Discrete transforms, convolutions and recurrences modules are implemented for SymPy, a computer algebra system (CAS) written purely in Python.</p>

The project aims to use Vision Transformer-based architectures to classify high-energy particles. The data consists of multi-channel simulated images from the calorimeter itself. The photos are used to classify the collision events. Previously various CNN-based architectures were built for the same, but in this project, we will try to use Visual Image Transformer-based architectures to classify the high-energy particle showers. We will experiment with various ViT-based architectures to find the highest possible validation metrics. We will also experiment with various advanced techniques used by contemporary deep-learning researchers to give a bleeding edge to the project. ViT is a much more powerful algorithm in comparison to the vanilla CNNs and if trained properly with a large amount of data often performs much better compared to CNNs. The main outcome of this project will be a Transformer based image classification architecture than can outperform CNNs in the classification of high energy particle images.

<p>The Vorto Web based IoT System Modeling tool provides user to design/model the IoT solution in a graphical way using web browser (Chrome/Firefox) using HTML5 canvas.</p>

<p>This is a user interface to draw and configure deep neural networks and supports import / export of model configuration file from / to caffe &amp; tensorflow deep learning frameworks</p>

<p>The goal of this project is to replicate research in Hierarchical Hidden Markov Models (HHMM) applied to financial data. This model is a generalization of Hidden Markov Models (HMM), which in turn is part of the Dynamic Bayesian Networks (DBN) family. I identified three academic works with interesting ideas and application that do not provide data and code. Replication will allow future readers to assess the credibility of the results and program improvements on top of fully working code. Ultimately, published code can be integrated into current research code (for example, for backtesting) and will be proposed as a <a href="http://mc-stan.org/documentation/case-studies" target="_blank">Case Study</a> to the Stan Development Team.</p> <p>The concept of hidden states could enrich many trading strategies. A detailed replication that provides literature review, reproducible code and solid documentation will allow future readers to implement HHMM logic into existing trading frameworks (for example for covariates, signals and benchmarks). Published code for this GSoC 2017 may be leveraged in future work to provide a generic implementation of hidden state models to already existing R Packages for trading analysis.</p>

The component-model proposal merges and supersedes the Module Linking and Interface Types proposals. With this feature, WasmEdge can execute multiple modules wasm with Module Linking and and more flexible types with Interface Type.

The emergence of blockchain interoperability is reducing the risk of investing in blockchain by avoiding vendor lock-in, leveraging interoperation with off-chain systems, and providing a truly open ecosystem, enabling a network of blockchains. In particular, in the technical report we wrote, we designed a survey aimed to understand the socio-technical challenges that blockchain interoperability poses, with the goal of offering a unique opportunity to link the underlying technology with human experience and values. We discovered that end-users are particularly concerned with visualizing and analyzing cross-chain cost and throughput, with some indication that measuring energetic expenditure in the cross-chain setting would also be valuable. Following that survey, we implemented a simple PoC that obtains transaction receipts from different blockchains and aggregates them in a unified repository. This project aims to build on top of what was done and create a cross-chain model (taking into account what end-users what to analyze from such model) automatically from input cross-chain transactions. This project will take a cross-chain use case, create a model and allow end-users to extract relevant metrics, promoting a better understatement of the business logic running on blockchain infrastructures, and of interoperability.

<p>TensorBoard is a TensorFlow tool for models visualization. It can generate histograms, graph and help debug and improve the neural network. To help the community, this proposal aims to produce code examples, guides, and tutorials to expose the advantages of TensorBoard and show how to extract relevant information about neural networks with the tool.</p>

FluxML is a deep learning stack built in 100% Julia code, which aims to provide high-speed and light-weight abstraction of GPU, auto-differentiation and deep learning model composites. This project will introduce a specific benchmarking system to adapt to the current requirements of the FluxML community. The system includes GitHub Actions workflows that are triggered by specific events, a FluxML-specific benchmarking tool and a repository for benchmarking code.

The motivation behind this project is to extend and introduce a fully functional MCP server for pgmoneta. To achieve this ultimate goal, we have identified a set of subgoals that will transform the current implementation into a production ready system: Introduce Non-Blocking I/O for Intensive I/O Operations (e.g Backup). Architectural Abstraction refactoring via the Rig Framework. Production-Level Security & Authentication. Knowledge Retrieval via MCP Resources. Testing and Documentation

This project aims to improve the handling of transposed diffuse irradiance and optical loss modeling in the pvlib-python package. Currently, pvlib’s transposition models provide inconsistent output, with some providing only total sky diffuse radiation (e.g. isotropic), while others allow for outputting the isotropic, circumsolar, and horizon components. Due to this, the ModelChain workflow only allows users to define a single angle of incidence (AOI) model, and then applies the resulting incident angle modifier (IAM) to total irradiance, limiting physical accuracy. The proposed GSoC project will start by standardizing the outputs of diffuse transposition models already implemented in pvlib, introducing a consistent interface to return circumsolar, isotropic, and horizon components. Building on this, the project will extend ModelChain to support component-specific IAM calculations, enabling more accurate modeling of optical losses by distinguishing between direct and diffuse contributions. These enhancements will improve both the flexibility and accuracy of pvlib simulations while maintaining full backward compatibility. The project will include comprehensive testing and documentation to ensure reliability and ease of use.

This project explores how lightweight Gemma models can be applied in life-critical situations through an open-source Crisis Response Toolkit. The main deliverable, One Minute Agent, is an offline AI assistant designed to provide first-aid guidance during emergencies, bridging the gap before responders arrive and helping improve survival rates. The work also investigates offline deployment, function-calling abilities, and agent workflows to demonstrate how Gemma can support real-time decision-making in crisis scenarios.

<p>Scalable congestion controls such as DCTCP improve performance over Reno and Cubic, which perform badly in high-speed networks (because of their slow response with large congestion windows). Several additional modifications over DCTCP have been drafted into the protocol called TCP Prague, that aims to integrate scalable congestion control into the Internet while still allowing it to coexist with current Classic protocols. This project would complete the integration and testing of fallback detection, RTT independence and pacing into the TCP Prague model of ns-3. The project would also validate the aforementioned implemenation against Linux and document the changes made.</p>

<p>A wide variety of problems in scientific computing and data analysis need to compute accurate shortest paths and distances along curved domains (e.g., curved surfaces representing the boundary of three-dimensional objects). The Computational Geometry Algorithms Library (CGAL) is a broadly-used open source framework for a wide variety of problems in geometric data analysis, but currently does not incorporate state-of-the-art methods for distance transforms. The goal of this project is to implement and integrate such methods into CGAL, specifically using methods as described in the paper "The Heat Method for Distance Computation" by Crane et al.</p>

<p>The objective of the problem arises from the reconstruction of gene genealogies problem - from a given genotypes of sequence data from contemporary individuals and an extended pedigree of genealogical relationships among them we have to decide and be very clear about what we should consider to be nodes information in genealogy tree according to kinds of dependencies among these nodes which should be considered to be edges of the tree. This project will focus on the development of novel algorithms for advanced reconstruction of the gene genealogy by performing inference on graphical models, specifically, implementing the greedy approach and loopy belief propagation algorithm on the succinct tree sequences. The implemented algorithms would undoubtedly be of use to integrate into the existing tskit and msprime libraries.</p>

<p>The goal of this project is to provide users of the mlr package with a way to visualize what happens during the tuning process that identifies the best hyperparameters for a learner given the data. Looking at popular packages in the machine learning community such as sklearn or caret, the model evaluation visualization process either requires lots of boilerplate code or is sparsely implemented across packages. This project will rectify the situation for R users of the mlr package by implementing three new features: plotting of a single hyperparameter against a scoring function, plotting multiple hyperparameters and scoring functions, as well as support for ablation analysis. The design of the features will focus on ease of use for the end user and customization. Users can choose to leverage automatically generated plots or use the data in some other manner downstream.</p>

The project Scrum Helper is designed to aid individuals in generating and injecting daily scrums into popular email clients such as Gmail, Outlook, and Yahoo. This initiative focuses on automating repetitive tasks involved in daily scrum reporting by leveraging user activity data from platforms like GitHub. The goal is to streamline scrum generation, improve developer productivity, and promote efficient communication within teams. Key Features: 1.Cross-Platform Support: Generate scrums in Gmail, Outlook, Yahoo, or via a standalone popup. 2.GitHub Integration: Auto-fetch activity to pre-fill scrum content. 3.Customizable & Shareable: Preview, edit, copy, and save scrum templates. 4.Secure & Optimized: OAuth-based access, responsive UI, and error handling.

<p><code>NicheToolBox</code> project will be an R package with a friendly Graphical User Interface (GUI) developed using <code>shiny</code> framework that aims to facilitate the process of building niche models and estimate the species distributions. To do the above it will incorporate functions to curate species occurrence data (clean duplicated records) and build models to estimate species niches (Bioclim, MaxEnt, Ellipsoid model) and distributions. After building a model the user will have the chance to evaluate its performance using Partial Roc, Confusion matrices and the associated metrics to it. Finally in order to make the process of niche modeling transparent, the application will have an option to download a workflow (in html, pdf and .doc) with the code that reproduces all the analysis that the user has made inside the application; this workflow can be shared with users interested to learn how to make a niche model using the R language.</p>

This proposal aims to contribute to addressing the challenge of generating formal SPARQL queries from natural language questions, thereby facilitating intuitive question-answering from open knowledge graphs, such as DBpedia. Among the three major approaches to this task, i.e., classification, ranking, and translation [Chakraborty et.al, 2019], this proposal will primarily focus on the latter one (translation). The key motivation behind choosing the translation approach is the significant performance improvement observed with the use of pre-trained language models such as T5 or GPT models in translation tasks.

Based on existing datasets, the issue aims to build an advanced benchmark for edge-oriented domain-specific large models on KubeEdge-Ianvs. It aims to help all Edge AI application developers validate and select the best-matched domain-specific large models. For Edge AI service providers, it also helps identify which scenarios, edge nodes, or even locations could have the best performance or improvement for their models. Expected Outcome: 1. Domain-specific Large Model Benchmark for the edge, including test datasets, testing toolkits, and usage guidelines. 2. (Advanced) Design and implementation of specific evaluation metrics. 3. (Advanced) Survey and research reports.